Artificial intelligence officially gets scary
- Thread starter Leo
- Start date
Murphy
cat malogen
Article that seems to confirm current suggestions that the Dall-E image maker has developed its own secret language without any prompting
hothardware.com
They've denied it but there's definitely something to this
Apoploe vesrreaitais
An Image Generation AI Created Its Own Secret Language But Skynet Says No Worries
A researcher claims that DALL-E, an OpenAI system that creates images from textual descriptions, is making up its own language.
They've denied it but there's definitely something to this
Apoploe vesrreaitais
Clinamenic
Binary & Tweed
I wonder how much of these are just artifacts of the training sets? Even if thats the case, it's still effectively sensing patterns that don't make sense to us.Article that seems to confirm current suggestions that the Dall-E image maker has developed its own secret language without any prompting
An Image Generation AI Created Its Own Secret Language But Skynet Says No Worries
A researcher claims that DALL-E, an OpenAI system that creates images from textual descriptions, is making up its own language.
They've denied it but there's definitely something to this
Apoploe vesrreaitais
If you read this, where they've had access to the API etc, it says that the training set for DALLE was CLIP, so definitely not directly, though this bird phrase does seem to have a bit of a correlation
The vegetable one is weird as well, no doubt they'll find more as they dig. Did anyone read Life 3.0?
vimothy
yurp
I imagine thats exactly what's happening. if you think of image recognition, the intermediate layers in a neural net build up abstract images which are then used by later layers to match against more complicated and concrete images. you could potentially feed the network an image based on those abstract images and have it recognise it as a cat or whatever, even though it's completely abstract and synthetic
Slothrop
Tight but Polite
Hack the planet:
spectrum.ieee.org
Slight Street Sign Modifications Can Completely Fool Machine Learning Algorithms
Minor changes to street sign graphics can fool machine learning algorithms into thinking the signs say something completely different
spectrum.ieee.org
Corpsey
bandz ahoy
Interesting how there's this gen-z/ultraonline/4chan humour that's almost replicating what a subpar AI might post as a joke
Mr. Tea
Let's Talk About Ceps
Who could possibly have seen that coming, eh?
Mr. Tea
Let's Talk About Ceps
Mancbook will never be as good as Scousebook.
wektor
Well-known member
The key takeaway from the Washington Post story isnt necessarily "sentience," which is an absurd idea in itself- the real issue would always be 'intelligence' above a certain level and how it is utilised, and by whom- but rather the furore being caused by individuals in the project itself
Google may try and paint this lad as a crank- that seems to be the attempt so far- but this will be weaponised with 18 months as part of a luddite post-hyperinflation culture war.
Watch the narrative
Google may try and paint this lad as a crank- that seems to be the attempt so far- but this will be weaponised with 18 months as part of a luddite post-hyperinflation culture war.
Watch the narrative